t,d deletion", or "(t,d)" has been the object of variationist studies for over half a century and continues to play a key part in theoretical debates about phonological representations both within and beyond variationist linguistics.Whereas they differ in their interpretations of the findings, most studies of the variable share a set of key assumptions about the nature of the variation involved.This paper addresses some of those assumptions, demonstrating how an examination of the detailed phonetics of the data raises fundamental problems which suggest that, in the absence of independent evidence to the contrary, (t,d) is best modelled as a Connected Speech Process, albeit a cognitively determined one.